Agreed, but it sounds like a bug to me. It would be much faster to read the row given the rowid and then use whatever column values(s) come up in indexes to get rid of the index entry. I can sort of understand a non-unique index maybe using an index ffs, but a unique one? Yet, stranger things have been seen... I suppose it'd also depend on the index depth as well? Plenty of ideas for some in-depth investigation, I reckon! -- Cheers Nuno Souto in wet Sydney, Australia dbvision@xxxxxxxxxxxx rjamya wrote,on my timestamp of 5/10/2009 9:26 PM:
I think you are right ... when delete happens on a (indexed) column value, the index deletion should be quick but since rowid is used, a index ffs is probably the fastest oracle can do.On Mon, Oct 5, 2009 at 7:10 AM, <troach@xxxxxxxxx <mailto:troach@xxxxxxxxx>> wrote:Since a btree index is organized by columns, not rowid (assuming its a btree index) oracle need to full scan it for that rowid. Since rowid points to the row in the table and not the index, it has no choice but to full scan the index. Someone please correct me if I'm wrong?
-- //www.freelists.org/webpage/oracle-l